The Mystery of Cell Processors

LucidBeast writes "Consumer appliances requiring more computing power Sony, IBM and Toshiba started 2001 developing "Cell"-processor that comprises of multiple processor cores and should give performance ten times of conventional processors. Now the CNN Money reports that details of the processor will be released Feb. 6-10 at the International Solid State Circuits Conference in San Francisco. Also reported by EE Times. Rumors also tell that Sonys PS3 development platform has already been shipped to some developers equipped with the cell processor."

Comments

What they didn't mention

[nick-less]

details of the processor will be released Feb. 6-10

it gives a 10 times performance gain over a normal processor, from the year 2001 of course, which will be something like a 1.3 GHz P4 or a 800 MHz Celeron, both introduced in january 2001 ;-)

What language do they use?

[Realistic_Dragon]

These multi core and multi processor systems can be a bugger to program for because handling concurrrency in a way that doesnt cause deadlocking is a major pain in the ass.

One of the better ways is to model out the program in CSP (or a variant thereof) and then write in a specially designed language like Occam (developed for the original transputer, but ported now to x86). These give you code that cannot deadlock or livelock or suffer from resource starvation without needing any of the complex and buggy hacks you see in things like the Linux kernel. And the Linux kernel only has to deal with a few processors... scalling to a few thousand processors in C would require a programmer of insane genius or the implimentation of effectivly a new language on top of C to handle the problems caused.

So, what language do developers use to target this? Is it something elegant designed for the problem at hand?

Will be interesting...

[TommyBear]

I currently work at a game studio here in Melbourne Australia and we're looking at next gen stuff (currently we develop xbox, ps2, PC games). Anyway, today at a meeting, one of the senior developers told our group that 4 had been selected to go to a little show and tell by IBM/Sony in Melbourne, where some of the secrets of the "Cell" processor would be demonstrated/explained to the group. Apparently we were only able to get 4 spots at this event.

So I'm exicited looks like the tech in just around the corner and so are the multi-core platforms (like XBOX2 and PS3).... yay!

Basicly, it is marketspeak...

[Kjella]

...the current computer architecture is nearing its limits yes, but it has no relationship to the content. A modern processor is very well capable of decoding HDTV content, probably encode too if you can accept less than super compression.

Of course, I see where it is going, I assume these Cell chips will be used to control hardware encoders/decoders with hard real-time limits (i.e. no frame skips and such crap). Taking the best of "dumb" hardware players of today, combined with the multitasking and flexibility of general computers.

But it is still a computer in drag. If anything, this seems more like a "retro" trend of the past, when you had active NICs/HDD controllers/whatnot with processors of their own. Now it is back with Cells instead. Just like terminals, we're coming full circle.

Kjella

Intel Prescott vs. Motorola 68k

[G4from128k]

This does not surprise me in the least. A Prescott processor has 125 million transistors, a Motorola 68000 had 68000 transistors. Yet the Prescott is not 1838 times more productive on a per clock-cycle basis. Admittedly, some of those Prescott transistors go to cache, superscalar magic, creating long fast pipes to achieve the GHz and implementing nifty MMX features. Even so, fabbing a 68k in 90 nm would create a tight little processor that is not 1800 times slower than the Prescott.

Thus, one can imagine creating a tighter core processor design with a budget of a million transistors each (15 times the original 68k budget) with a few million for L1 cache and another million for glue and then place 20 of them on a single die. Add optical interconnects and that new optical-to-silicon technology invented recently (for multiple channels of GHz I/O to feed all those cores) and you have yourself a powerful little processor.

The point is that with a budget of 125 million transistors, designers can do more than create a bloated single-core CISC processor.

DRM For the Masses

[rsmith-mac]

After reading that press release, and correct me if I'm wrong, I'm not sure what's really "new" about the Cell other than On-chip hardware in support of security system for intellectual property protection. There are other Power designs already that do multicore, do high performance, and do vector ops(Altivect), so the only thing that I haven't heard about a design for is their security system.

Considering the companies involved, and the devices that they want to put the chip in, I'm really tempted to say that the Cell is nothing more than the biggest effort we've ever seen to get a DRM (trusted computing) CPU and associated parts on to the market. Obviously, this scares the bejesus out of me, since it would mean that these Cell devices would effectively be mod-proof; systems like Xbox Live already keep cheaters away, so this seems to be an attempt to stop modding alltogether. So, I have to ask: how is this going to benefit me, the consumer? If Live already gets rid of possible cheaters, how does stopping me from modding my box altogether help me?

If these assumptions are right, I don't like where this is going.